In the recent years Image Annotation has picked up a significance in the process of searching, retrieving and applying labels for images.The project involves in identifying and recognizing the objects within the given image using predefined neural network learning algorithms and tools. The model also includes recognition of complex images using partially annotated datasets for improved annotation process.
The model accomplishes two major tasks: Initially, to explain with labels and later to segment the picture with the comparable area, such that the expectation of the overall model will be useful in computer vision. The proposed model is carried out using MatLab and ImageNet dataset. The performance of the system demonstrates the presence of correct- ness within the inputted image using labels (tags) and partitioning pictures from different classes portraying to various scenes.
With the development of the Internet and digital imaging devices many large im- age collections are being created. Popular online photo-sharing sites like Flickr contain hundreds of millions of diverse pictures. Many organizations, e.g. li- braries, hospitals, governments and commerce have also been creating their large image databases by scanning paintings, manuscripts, prints and drawings. Search- ing and finding large numbers of images from a database is a challenging problem. Search engine they do not really capture the semantics or meaning of images well. For image retrieval systems based on text queries, the key problem is how to get the metadata such as captions, titles or transcriptions. Manual annotation is not practical for large volumes of image sets. Commercial image search engines for the Internet, e.g. Google image and Yahoo image, use the text surrounding each image as its description. However, these search engines entirely ignore the visual content of the images and the surrounding text doesnt always relate to the visual content of an image. The consequence is that the returned images may be entirely unrelated to users needs.
Vision is the richest sense that a human being has which computer does not have and will consist of a tedious process to achieve the same for a computer. Object recognition and classification play a major role in this field.So we need a framework which is used for detecting objects from the image and annotate them with the proper tags, which will be used for problems stated above.
Image annotation is a complex job of detecting objects and classifying each ob- jects in a given image. Even though the process is extremely useful in some cases, the complexity of the process limits many novice developers from using image annotation and object detection in their projects. So we are developing a user friendly framework that will do simplest of the image annotation tasks and help novice developers in their projects that might need object detection and classification.
Here we have used MatLab Computer Vision toolbox which is efficient and faster than OpenCV. Objective is, we are trying to extract feature from an image and annotate them with appropriate tags for the object present in them. We extracted features and classify the image using AlexNet ( A pre-trained convolutional Neural Network) which uses SIFT(Scale-invariant feature transform) which is advanced method than ORB and SURF, which helps in reducing noise. And also it uses dropout technique to selectively ignore single neurons during training avoiding the averaging effects of average pooling. Benefits of proposed work are:
- Firstly, language selection is very important, MatLab is very efficient in terms of space and time.
- Secondly, we used AlexNet which is one of top most ImageNet Classification in the field of neural Network